Scholarlink

<p>For these flexible material applications, including soft robotics, wearable devices, and biomedical implants constructed from polymers, elastomers, and biological tissues, precise predictions of the deformation and failure behaviors are most desired. Traditional assessment for these has relied highly on experimental analyses, which are always costly and time-consuming. This work outlines a machine learning-based methodology for predicting material deformation/fracture characteristics based on microstructure image data. We trained and validated several machine learning models on a dataset of 15,000 images categorized into five distinct classes, each representing a different stage of material deformation. By applying transfer learning and pre-trained models from five established machine learning architectures, including ResNet50, we achieved a classification accuracy of 99.81% with a processing time of 2.9&#xa0;h. In other words, our method enables substantial reductions in complex experimental arrangements while introducing a scalable data-driven solution for material characterizations. The results make it clear that the ML models effectively capture the material property-deformation behavior relationship, promising high potential for future materials science and engineering applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

The use of machine learning algorithms to determine the rate of material deformation: a new approach

  • 期刊论文

  • 发表日期：

    2025年1月21日

查看全文
  • Feng Li,
  • Ying Chen,
  • Ruobing Li,
  • Xiaole Zhang,
  • Lanzhen Li

摘要

For these flexible material applications, including soft robotics, wearable devices, and biomedical implants constructed from polymers, elastomers, and biological tissues, precise predictions of the deformation and failure behaviors are most desired. Traditional assessment for these has relied highly on experimental analyses, which are always costly and time-consuming. This work outlines a machine learning-based methodology for predicting material deformation/fracture characteristics based on microstructure image data. We trained and validated several machine learning models on a dataset of 15,000 images categorized into five distinct classes, each representing a different stage of material deformation. By applying transfer learning and pre-trained models from five established machine learning architectures, including ResNet50, we achieved a classification accuracy of 99.81% with a processing time of 2.9 h. In other words, our method enables substantial reductions in complex experimental arrangements while introducing a scalable data-driven solution for material characterizations. The results make it clear that the ML models effectively capture the material property-deformation behavior relationship, promising high potential for future materials science and engineering applications.